Key facts
| Agent stack | Chat completions, function calling and embeddings are live |
| Use cases | Logistics queries, maintenance documentation, training drafts, policy retrieval |
| Human authorisation | No autonomous action; authorised personnel decide and act |
| Deployment | On-prem and air-gapped options for isolated networks |
| Tools | Read-only access scoped per system and role |
| Models | 30+ models behind one compatible API |
| Pricing model | Flat monthly self-serve plans with unlimited fair-use usage |
| Roadmap | Batch and assistants endpoints are coming soon |
TL;DR
- Apply agents to unclassified knowledge work, not command or control.
- Keep every action behind human authorisation and existing process.
- Run on-prem or air-gapped where networks are isolated.
- Scope tools and indexes per role so access matches authorisation.
- Measure documentation time saved and retrieval accuracy.
How it works, step by step
- Select one knowledge workflow such as maintenance documentation or logistics lookup.
- Confirm the deployment is on-prem or air-gapped and that no external traffic is required.
- Index approved manuals, policies and training material with revision metadata.
- Expose read-only tools scoped to the user's role and authorisation level.
- Require citations and route every decision through the existing chain of command.
- Log tool calls and model versions, then evaluate accuracy and time saved per task.
Try it yourself
Open the tool registry builder →
Where defense knowledge work benefits
On isolated networks, the value comes from retrieving and drafting rather than automating:
- Logistics queries: find stock levels, routing rules and supply references from approved systems.
- Maintenance documentation: retrieve procedures and summarise equipment history for technicians.
- Training material: draft lesson content and assessments from approved doctrinal sources.
- Policy retrieval: answer administrative and procedural questions with citations.
The agent supports the person accountable for the decision, and nothing more.
Architecture for isolated networks
Air-gapped environments impose constraints that shape the design:
- No external calls: models and retrieval run inside the environment, with updates delivered through your own process.
- Role-scoped tools: read-only access that matches the user's authorisation level.
- Revision-aware index: store revision and effective date so answers reflect current doctrine.
- Composer: outputs with citations and explicit assumptions.
- Authorisation gate: any action follows the existing chain of command.
Log tool calls and model versions for review and after-action analysis.
Data discipline and access control
Even inside an isolated network, classification and need-to-know rules apply. Segment indexes by classification and role, enforce least privilege on tool credentials, and keep prompts and logs within the same boundary. Use scoped keys, RBAC and audit logs to make access attributable, and pin model versions so results are reproducible during review. Where a programme permits only specific hardware or network zones, deploy on-prem or air-gapped accordingly. The institution's security policy remains the governing document; the platform provides the controls to implement it.
Evaluating in a closed environment
Prototype on plugsky-micro or plugsky-lite inside the target environment with a controlled document set, then use the 14-day full-access trial where policy allows to compare models on harder retrieval and drafting tasks. Track retrieval accuracy, citation correctness, stale-revision rate and time saved per task. Self-serve plans are flat monthly with fair-use usage, which keeps budgeting simple for steady workloads. Expand by work role, with versioned prompts and a re-run of the evaluation set whenever doctrine or manuals change.
Honest comparison
| Capability | Plugsky | Typical agent framework | Building in-house |
|---|---|---|---|
| Tool calling | Live function calling over role-scoped, read-only APIs | Framework-specific adapters | You build and maintain |
| Air-gapped option | On-prem and air-gapped deployment available | Usually cloud-only | You operate everything |
| Human control | No autonomous action; authorisation stays with people | Varies | You design controls |
| Pricing | Flat monthly self-serve, fair-use usage | Varies by framework and model | GPU plus operations cost |
| Model choice | 30+ models behind one API | Varies by provider | You host every model |
Frequently asked questions
Can the agent take autonomous action?
No. Keep agents advisory, with read-only tools and every decision routed through the existing chain of command and authorisation process.
Which endpoints do defense agents use?
Chat completions, function calling, JSON mode and embeddings are live. Batch and assistants endpoints are coming soon.
Can it run air-gapped?
Yes. On-prem and air-gapped deployment is available for isolated networks, using the same compatible API and agent code.
Is there a free plan?
Yes. The free plan includes plugsky-micro and plugsky-lite with no credit card, which suits a controlled prototype inside your environment.
How do we handle classification levels?
Segment indexes and tool credentials by classification and role, enforce least privilege, and keep prompts and logs inside the same boundary.
How does pricing work?
Self-serve plans are flat monthly with unlimited fair-use usage. See the live pricing page for current plans and enterprise options.
How should we measure it?
Track retrieval accuracy, citation correctness, stale-revision rate and documentation time saved per task.